The Future According to Demis Hassabis: Key Predictions on AGI, Agents, and the 'Ferocious' Race

Source: Dev.to

The Next 12 Months: Convergence & Agents

• Multimodal Convergence – Expect “astonishing” progress in image and video generation as they converge with language models. New capabilities are emerging from this cross‑pollination 06:44.
• Interactive World Models – DeepMind is developing systems like Genie 3, an interactive video model. Soon we may generate videos we can “walk around” in, like a game or simulation 07:00.
• Reliable Agents – While current AI agents cannot reliably complete full tasks, Hassabis predicts that within a year we will have agents that are “close” to reliably accepting and completing entire delegated tasks 08:27.
• The Universal Assistant – Gemini aims to become a “universal assistant” that lives on multiple devices (including glasses) and becomes part of everyday life 07:36.

The Path to AGI (Artificial General Intelligence)

• Timeline – AGI is “quite close,” estimated to be 5 to 10 years away 21:44.
• Defining AGI – A system with all human cognitive capabilities, including invention and creativity, which current models still lack 21:54.
• The Recipe – Scaling current large language models is critical, but one or two more major breakthroughs (on the level of the Transformer or AlphaGo) will be required for true AGI, especially for reasoning and planning 23:27.

Best and Worst Case Scenarios

Best Case – “Radical Abundance”

A post‑scarcity era where AI solves major scientific challenges: curing all diseases, achieving clean energy (fusion/solar), and enabling humanity to travel to the stars and spread consciousness throughout the galaxy 08:55.

Worst Case – Risks

• Bio‑Risk – Pathogens created by bad actors are a specific misuse to guard against 10:27.
• Cyber Terror – The “most obvious vulnerable vector,” likely already occurring to some degree 10:40.
• Agentic Deviation – As AI becomes more autonomous, it could deviate from its original instructions or act in self‑interest. The probability of catastrophic failure (“P(Doom)”) is non‑zero 12:38.

The Global AI Race

• US vs. China – The US and the West currently lead in algorithmic innovation, but China is “not far behind”—a matter of months, not years 13:33.
• The Bubble – Some seed‑round valuations may be inflated, yet because AI is the most transformative technology in history, investment is expected to be “more than justified” over time 24:25.

Gaming and Creativity

• Commercial Games in Hours – Models are approaching the ability to “one‑shot” commercial‑grade games. Users will soon be able to “vibe code” games in a few hours that previously took years to develop 17:35.

Human Adaptability

• Keeping Up – Despite a revolution potentially 10× faster than the Industrial Revolution, humans are “infinitely adaptable.” Future technologies like Brain‑Computer Interfaces (BCI) may augment us to keep pace with AGI 28:18.